An automatic identification algorithm for system order is proposed to effectively improve the efficiency of modal parameter identification of bridge structures. Modal parameter identification is one of the important links of bridge structure health monitoring，and the accuracy of identification will directly affect the state evaluation of bridge structures. However，as one of the commonly used algorithms in time domain modal analysis，the stochastic subspace algorithm has the defect of artificial participation in system order determination. Therefore，a new approach for integrating the singular value jumping method into "logarithmic-processing" is proposed to quickly determine the evidently jump point of the singular value，which solves the problem that the singular value jump method is difficult to determine the order of the higher-order system. The frequency confidence factor and mode confidence factor are constructed to determine the value range of the real orders. The confidence coefficient is constructed by linear weighted averages method to determine the real order of the structural system. The proposed algorithm is applied to a large cable-stayed bridge，and its identification results are compared with the field test values and the finite element results. The results show that the proposed algorithm can realize the automatic determination of the system order and the automatic identification of modal parameters，and its identification results have better accuracy and reliability.